Summary Under the proposed Cloud and AI Development Act (CADA), AI industrial innovation, physical AI, and public sector AI projects can access EU high-performance computing (HPC) resources through a coordinated effort between the Union and Member States. However, the legal obligation differs significantly from that of "frontier AI priority projects." While frontier projects benefit from a mandatory matching guarantee, industrial and public sector projects fall under Article 9(3), which requires the Union and Member States to endeavour to provide sufficient computing resources. This access is contingent on available capacity within the EuroHPC framework and is part of a broader strategy to reduce dependency on non-European infrastructure.

Detail

The Cloud and AI Development Act (CADA), as proposed in COM(2026) 502 final, establishes a tiered framework for allocating scarce high-performance computing (HPC) resources. The Act aims to strengthen Europe's technological sovereignty by ensuring that strategic AI capabilities are developed on European infrastructure. The mechanism for accessing these resources is not uniform; it depends entirely on the classification of the AI project.

To understand the pathway for industrial and public sector projects, one must first distinguish them from the highest tier of priority: "frontier AI priority projects."

The Strong Guarantee: Frontier AI Priority Projects

The Act creates a specific, high-priority category for "frontier AI priority projects" defined in Article 8. These are pioneering projects focused on scaling up frontier AI technologies, such as advanced cybersecurity models or next-generation multimodal systems. The criteria for designation are strict, requiring the project to be a pioneering effort involving broad participation from entities across the Union, often through European digital infrastructure consortia (EDICs).

For these designated projects, the Act imposes a robust, mandatory obligation on the Union and Member States. Article 9(1) states that the Union and Member States "shall ensure that sufficient AI computing resources from their compute capacities are allocated to support the development of frontier AI priority projects that fulfil the criteria set out in Article 8, within the limits of available capacity."

Crucially, Article 9(2) introduces a "matching" mechanism that acts as a powerful financial and resource incentive:

"The Union shall at least match the AI computing resources contributed by Member States to frontier AI priority projects to the extent that sufficient AI computing capacity is available within the Union's share of European high performance computing access time."

This creates a "1-for-1" guarantee (at minimum) for frontier projects. If a Member State contributes compute time, the Union is legally bound to match it, effectively doubling the available resources for these strategic assets, provided capacity exists within the EuroHPC share.

The Endeavour: Industrial, Physical, and Public Sector AI

For projects that do not meet the strict "frontier AI priority" criteria but remain critical to the EU's strategic autonomy, CADA provides a different, less binding mechanism. This covers three specific strategic domains:

  1. AI Industrial Innovation: Sector-specific AI models for industries such as healthcare, automotive, manufacturing, energy, and defence.
  2. Physical AI: AI systems capable of perceiving and acting upon the physical environment, including robotics, autonomous drones, and self-driving vehicles.
  3. Public Sector AI: AI models used to improve public service delivery, decision-making, and administrative efficiency across critical public domains.

The provision governing access for these projects is found in Article 9(3). The text states:

"The Union and the Member States shall endeavour to provide sufficient computing resource for AI industrial innovation, physical AI and public sector AI projects."

The use of the word "endeavour" is legally significant in EU legislative drafting. Unlike the mandatory "shall ensure" (Article 9(1)) and "shall at least match" (Article 9(2)) language used for frontier AI, "endeavour" indicates a strong political and administrative commitment to facilitate access, but it does not create an absolute, guaranteed entitlement to specific compute volumes.

This distinction implies that while the Commission and Member States must actively work to allocate resources, optimize existing infrastructure, and remove barriers to access, the final allocation remains subject to capacity constraints and prioritization decisions. It is a duty of means, not a duty of result. If capacity is exhausted by frontier projects or other strategic needs, industrial and public sector projects may face competition for remaining resources.

How Access Works in Practice

Access to these resources is not automatic and operates within a broader ecosystem of monitoring and national strategy.

  • Capacity Monitoring: The Commission is tasked with monitoring the Union's compute capacity, the volume of demand, and the size of the capacity gap under Article 15. This data informs the availability of resources for the "endeavour" obligation.
  • EuroHPC Integration: The compute resources largely derive from the European High Performance Computing (EuroHPC) Joint Undertaking. Recital 35 of the proposal clarifies that the EuroHPC JU access policy "should be accommodated to reflect the allocation of such computing resources in an efficient, transparent and timely manner without prejudice to the continuity of ongoing operations and the rights of projects already benefiting from allocated EuroHPC AI computing resources."
  • National Strategies: Member States are required to adopt national cloud and AI strategies within one year of the Regulation's entry into force (Article 7). These strategies must include measures to invest in high-intensity computing infrastructure, including AI factories and quantum computers. These national strategies will likely define the specific application processes, eligibility criteria, and allocation windows for industrial and public sector entities to request compute time under the Article 9(3) "endeavour" clause.

Strategic Context

This tiered approach reflects the EU's strategic prioritization. Frontier AI is viewed as a global competitive battleground where the EU risks falling behind, justifying a stronger, guaranteed resource commitment. In contrast, industrial, physical, and public sector AI are viewed as essential for competitiveness, societal benefit, and public order. While they warrant significant support, the "endeavour" clause allows for the flexibility needed to manage a potentially larger volume of applicants and varying levels of criticality across different Member States.

What this means for you

For CTOs, architects, public sector IT managers, and SMEs operating in the EU, the distinction between "frontier" and "industrial/public/physical" AI is crucial for resource planning and project scoping.

1. Manage Expectations on Guarantees If your project involves developing a foundational model that competes at the global frontier, you may qualify for the stronger matching guarantee under Article 9(2). However, if you are building a specialized AI for automotive defect detection (industrial), a warehouse robot (physical AI), or a public health triage system (public sector), you fall under Article 9(3). You should plan for access to compute as a facilitated opportunity rather than a guaranteed right. You will need to demonstrate the strategic value of your project to secure priority in allocation queues, as the Union is only obliged to endeavour to provide resources.

2. Engage with National Strategies Since Member States are tasked with adopting national cloud and AI strategies (Article 7), your primary point of contact for compute access will likely be national. Monitor your country's specific implementation of these strategies. National authorities will likely establish application windows or criteria for allocating the "endeavoured" compute resources. Early engagement with national digital innovation hubs, now refocused as "Centres for AI" under Article 5, can help position your project for favorable consideration.

3. Highlight Strategic Alignment To maximize your chances of securing compute under the "endeavour" clause, align your project with the EU's stated priorities. For industrial AI, emphasize supply chain security and competitiveness. For physical AI, highlight safety and autonomy in unstructured environments. For public sector AI, focus on efficiency, citizen trust, and the preservation of public order. The Commission and Member States are more likely to prioritize projects that clearly demonstrate a reduction in dependency on non-European providers or a significant boost to EU industrial sovereignty.

4. Prepare for Competitive Allocation Given that capacity is finite, expect a competitive landscape. The proposal mentions that the EuroHPC JU access policy will be adjusted to handle these allocations. This suggests a transparent, criteria-based system. Ensure your technical documentation is robust and that you can clearly articulate the compute requirements and expected outcomes of your project. Be prepared to show how your project contributes to the "grand challenges" outlined in Annex I, such as environmental sustainability or industrial AI adoption.

Common misconceptions

Misconception 1: "Endeavour" means no obligation. While "endeavour" is less binding than "shall ensure," it is not meaningless. In EU law, an obligation to "endeavour" requires the authorities to take all reasonable steps to achieve the goal. They must actively allocate resources, remove barriers, and prioritize these sectors. It creates a political and administrative pressure to deliver, even if it doesn't guarantee a specific teraflop count. It is a duty of diligence, not a duty of result.

Misconception 2: All AI projects get free compute. CADA does not provide free compute for all AI projects. It establishes a framework for allocating existing and newly built EU HPC capacity. Access is targeted at strategic initiatives that contribute to the EU's competitiveness and sovereignty. Small-scale or purely commercial projects without strategic industrial or public benefit may not qualify for these specific allocation streams.

Misconception 3: Only giant corporations can access these resources. The proposal explicitly aims to support SMEs and start-ups. The national strategies and Centres for AI are designed to help smaller entities navigate the application process and access compute. The tiered approach ensures that while frontier AI might be dominated by large players, industrial and public sector AI resources are also available to a broader range of innovators.

Misconception 4: CADA replaces the AI Act. CADA and the AI Act are complementary. The AI Act regulates the safety and fundamental rights implications of AI systems. CADA focuses on the infrastructure and compute capacity needed to develop and deploy them. Accessing compute under CADA does not exempt you from compliance with the AI Act's requirements for high-risk AI systems.

Official sources

Related

This is general information about a draft EU regulation, not legal advice.